Optimal Scheduling of Railway Traction Power Supply Systems with Microgrid Integration for Cost and Power Quality Improvements

Decarbonizing transport is essential to achieving net-zero greenhouse gas emissions, yet the sector remains heavily dependent on fossil fuels and is among the most difficult to decarbonize. Railway electrification offers a promising pathway because of the centralized and standardized nature of rail operations. However, its further expansion is constrained by the high cost and long lead time of railway power infrastructure and by the additional stress that highly variable traction demand can impose on already constrained power grids.

This thesis investigates railway microgrids as a means of addressing these challenges by integrating and coordinating distributed energy resources, including renewable generation and energy storage, within railway power supply systems. Combined with advanced energy management, railway microgrids can enhance local power supply capability, operational flexibility, and the ability of electrified railways to operate as active prosumers. Three progressive application scenarios are investigated.

First, an advanced machine-learning-based energy management framework is developed for a railway microgrid integrated at the feeder station. The microgrid coordinates the traction network, local renewable generation, the local power grid, hydrogen storage, and aggregated electric vehicle fleets to supply traction and non-traction loads. A bidirectional long short-term memory model is employed for renewable energy forecasting, while an improved Twins-Actor deep deterministic policy gradient method is developed for optimal scheduling. A fast recursive algorithm is further introduced to reduce the computational burden of the learning agents. Case studies show that the framework can reduce daily operating cost by up to 18.7%, while reducing agent size by up to 78.6% and computational cost by up to 93.1%.

Second, railway microgrid integration is extended to weak terminal nodes of the traction network, and a collaborative energy management framework is developed between terminal microgrids and co-phased feeder stations. A computational-cost-aware heuristic reinforcement learning method coordinates the scheduling of railway microgrids and feeder-station resources to accommodate increasing traction demand while mitigating three-phase voltage imbalance. Case studies demonstrate daily operating cost reductions of 15.1% in representative scenarios and 19.8% on average under real-world operating conditions, together with computation-time reductions of up to 81% compared with conventional approaches.

Third, the thesis extends railway microgrid energy management from economic scheduling to traction-network-level voltage-aware operation. A terminal-station railway microgrid integrating on-site energy storage, renewable generation, and the local utility grid is coordinated to improve economic performance while maintaining traction voltage levels. A hierarchical reinforcement learning framework is proposed, in which an upper-level agent dynamically adjusts allowable voltage reference ranges and a lower-level agent performs energy scheduling under these references. Case studies show that the proposed framework can reduce daily operating costs by up to 35.2% compared with the case without railway microgrid integration while improving the voltage profile at weak traction-network nodes.

Overall, this thesis establishes a progressive framework for microgrid-enabled railway electrification, from multi-energy coordinated scheduling at feeder stations to coordinated terminal-station support and voltage-aware operation. The results demonstrate that railway microgrids, combined with advanced energy management and optimization methods, can improve cost-efficiency, operational flexibility, and power quality while reducing the impact of increasing traction demand on the wider power system. These findings provide practical insights for supporting the continued expansion of railway electrification and transport decarbonization.